BOSO

BOSO performs bilevel optimization-based feature selection for linear regression to identify predictive variables in high-dimensional biomedical datasets.

Key Features:

• R package implementation: Implemented as an R package for computational use.
• Bilevel optimization: Uses bilevel optimization that optimizes an upper-level objective (model performance) and a lower-level objective (feature selection).
• Linear regression focus: Targets feature selection specifically within linear regression frameworks.
• High-dimensional data handling: Designed to address feature selection challenges in high-dimensional biomedical datasets.
• Benchmarking validation: Evaluated in comparative benchmarking against other prominent algorithms, reporting superior accuracy in feature selection tasks.
• Drug-sensitivity application: Applied to predict methotrexate response in cancer cell lines and identify features predictive of treatment response.

Scientific Applications:

• Drug sensitivity prediction (methotrexate): Used to analyze methotrexate response across cancer cell lines to identify predictive molecular features.
• Feature selection in biomedical studies: Selects predictive variables for linear regression models in complex, high-dimensional biomedical datasets.
• Informing treatment response analyses: Supports identification of features that may inform personalized therapeutic strategies in cancer research.

Methodology:

BOSO employs bilevel optimization that jointly optimizes an upper-level objective targeting model performance and a lower-level objective performing feature selection within linear regression models, implemented as an R package and validated via benchmarking against other algorithms.